import binascii import logging.handlers import mimetypes import os import base58 import requests import urllib import json import textwrap import random import signal from copy import deepcopy from twisted.web import server from twisted.internet import defer, threads, error, reactor from twisted.internet.task import LoopingCall from twisted.python.failure import Failure from lbryschema.claim import ClaimDict from lbryschema.uri import parse_lbry_uri from lbryschema.error import URIParseError from lbryschema.validator import validate_claim_id from lbryschema.address import decode_address # TODO: importing this when internet is disabled raises a socket.gaierror from lbrynet.core.system_info import get_lbrynet_version from lbrynet import conf, analytics from lbrynet.conf import LBRYCRD_WALLET, LBRYUM_WALLET, PTC_WALLET from lbrynet.reflector import reupload from lbrynet.reflector import ServerFactory as reflector_server_factory from lbrynet.core.log_support import configure_loggly_handler from lbrynet.lbry_file.client.EncryptedFileDownloader import EncryptedFileSaverFactory from lbrynet.lbry_file.client.EncryptedFileOptions import add_lbry_file_to_sd_identifier from lbrynet.lbry_file.EncryptedFileMetadataManager import DBEncryptedFileMetadataManager from lbrynet.lbry_file.StreamDescriptor import EncryptedFileStreamType from lbrynet.file_manager.EncryptedFileManager import EncryptedFileManager from lbrynet.daemon.Downloader import GetStream from lbrynet.daemon.Publisher import Publisher from lbrynet.daemon.ExchangeRateManager import ExchangeRateManager from lbrynet.daemon.auth.server import AuthJSONRPCServer from lbrynet.core.PaymentRateManager import OnlyFreePaymentsManager from lbrynet.core import utils, system_info from lbrynet.core.StreamDescriptor import StreamDescriptorIdentifier, download_sd_blob from lbrynet.core.Session import Session from lbrynet.core.Wallet import LBRYumWallet, SqliteStorage, ClaimOutpoint from lbrynet.core.looping_call_manager import LoopingCallManager from lbrynet.core.server.BlobRequestHandler import BlobRequestHandlerFactory from lbrynet.core.server.ServerProtocol import ServerProtocolFactory from lbrynet.core.Error import InsufficientFundsError, UnknownNameError, NoSuchSDHash from lbrynet.core.Error import NoSuchStreamHash, DownloadDataTimeout, DownloadSDTimeout from lbrynet.core.Error import NullFundsError, NegativeFundsError log = logging.getLogger(__name__) INITIALIZING_CODE = 'initializing' LOADING_DB_CODE = 'loading_db' LOADING_WALLET_CODE = 'loading_wallet' LOADING_FILE_MANAGER_CODE = 'loading_file_manager' LOADING_SERVER_CODE = 'loading_server' STARTED_CODE = 'started' WAITING_FOR_FIRST_RUN_CREDITS = 'waiting_for_credits' STARTUP_STAGES = [ (INITIALIZING_CODE, 'Initializing'), (LOADING_DB_CODE, 'Loading databases'), (LOADING_WALLET_CODE, 'Catching up with the blockchain'), (LOADING_FILE_MANAGER_CODE, 'Setting up file manager'), (LOADING_SERVER_CODE, 'Starting lbrynet'), (STARTED_CODE, 'Started lbrynet'), (WAITING_FOR_FIRST_RUN_CREDITS, 'Waiting for first run credits'), ] # TODO: make this consistent with the stages in Downloader.py DOWNLOAD_METADATA_CODE = 'downloading_metadata' DOWNLOAD_TIMEOUT_CODE = 'timeout' DOWNLOAD_RUNNING_CODE = 'running' DOWNLOAD_STOPPED_CODE = 'stopped' STREAM_STAGES = [ (INITIALIZING_CODE, 'Initializing'), (DOWNLOAD_METADATA_CODE, 'Downloading metadata'), (DOWNLOAD_RUNNING_CODE, 'Started %s, got %s/%s blobs, stream status: %s'), (DOWNLOAD_STOPPED_CODE, 'Paused stream'), (DOWNLOAD_TIMEOUT_CODE, 'Stream timed out') ] CONNECTION_STATUS_CONNECTED = 'connected' CONNECTION_STATUS_NETWORK = 'network_connection' CONNECTION_MESSAGES = { CONNECTION_STATUS_CONNECTED: 'No connection problems detected', CONNECTION_STATUS_NETWORK: "Your internet connection appears to have been interrupted", } SHORT_ID_LEN = 20 class IterableContainer(object): def __iter__(self): for attr in dir(self): if not attr.startswith("_"): yield getattr(self, attr) def __contains__(self, item): for attr in self: if item == attr: return True return False class Checker(object): """The looping calls the daemon runs""" INTERNET_CONNECTION = 'internet_connection_checker' CONNECTION_STATUS = 'connection_status_checker' class _FileID(IterableContainer): """The different ways a file can be identified""" NAME = 'name' SD_HASH = 'sd_hash' FILE_NAME = 'file_name' STREAM_HASH = 'stream_hash' CLAIM_ID = "claim_id" OUTPOINT = "outpoint" ROWID = "rowid" FileID = _FileID() # TODO add login credentials in a conf file # TODO alert if your copy of a lbry file is out of date with the name record class NoValidSearch(Exception): pass class CheckInternetConnection(object): def __init__(self, daemon): self.daemon = daemon def __call__(self): self.daemon.connected_to_internet = utils.check_connection() class AlwaysSend(object): def __init__(self, value_generator, *args, **kwargs): self.value_generator = value_generator self.args = args self.kwargs = kwargs def __call__(self): d = defer.maybeDeferred(self.value_generator, *self.args, **self.kwargs) d.addCallback(lambda v: (True, v)) return d # If an instance has a lot of blobs, this call might get very expensive. # For reflector, with 50k blobs, it definitely has an impact on the first run # But doesn't seem to impact performance after that. @defer.inlineCallbacks def calculate_available_blob_size(blob_manager): blob_hashes = yield blob_manager.get_all_verified_blobs() blobs = yield defer.DeferredList([blob_manager.get_blob(b) for b in blob_hashes]) defer.returnValue(sum(b.length for success, b in blobs if success and b.length)) class Daemon(AuthJSONRPCServer): """ LBRYnet daemon, a jsonrpc interface to lbry functions """ allowed_during_startup = [ 'daemon_stop', 'status', 'version', ] def __init__(self, analytics_manager): AuthJSONRPCServer.__init__(self, conf.settings['use_auth_http']) self.db_dir = conf.settings['data_dir'] self.download_directory = conf.settings['download_directory'] if conf.settings['BLOBFILES_DIR'] == "blobfiles": self.blobfile_dir = os.path.join(self.db_dir, "blobfiles") else: log.info("Using non-default blobfiles directory: %s", conf.settings['BLOBFILES_DIR']) self.blobfile_dir = conf.settings['BLOBFILES_DIR'] self.data_rate = conf.settings['data_rate'] self.max_key_fee = conf.settings['max_key_fee'] self.disable_max_key_fee = conf.settings['disable_max_key_fee'] self.download_timeout = conf.settings['download_timeout'] self.run_reflector_server = conf.settings['run_reflector_server'] self.wallet_type = conf.settings['wallet'] self.delete_blobs_on_remove = conf.settings['delete_blobs_on_remove'] self.peer_port = conf.settings['peer_port'] self.reflector_port = conf.settings['reflector_port'] self.dht_node_port = conf.settings['dht_node_port'] self.use_upnp = conf.settings['use_upnp'] self.startup_status = STARTUP_STAGES[0] self.connected_to_internet = True self.connection_status_code = None self.platform = None self.current_db_revision = 4 self.db_revision_file = conf.settings.get_db_revision_filename() self.session = None self.uploaded_temp_files = [] self._session_id = conf.settings.get_session_id() # TODO: this should probably be passed into the daemon, or # possibly have the entire log upload functionality taken out # of the daemon, but I don't want to deal with that now self.analytics_manager = analytics_manager self.node_id = conf.settings.node_id self.wallet_user = None self.wallet_password = None self.query_handlers = {} self.waiting_on = {} self.streams = {} self.exchange_rate_manager = ExchangeRateManager() calls = { Checker.INTERNET_CONNECTION: LoopingCall(CheckInternetConnection(self)), Checker.CONNECTION_STATUS: LoopingCall(self._update_connection_status), } self.looping_call_manager = LoopingCallManager(calls) self.sd_identifier = StreamDescriptorIdentifier() self.stream_info_manager = None self.lbry_file_manager = None @defer.inlineCallbacks def setup(self): reactor.addSystemEventTrigger('before', 'shutdown', self._shutdown) configure_loggly_handler() @defer.inlineCallbacks def _announce_startup(): def _announce(): self.announced_startup = True self.startup_status = STARTUP_STAGES[5] log.info("Started lbrynet-daemon") log.info("%i blobs in manager", len(self.session.blob_manager.blobs)) yield self.session.blob_manager.get_all_verified_blobs() yield _announce() log.info("Starting lbrynet-daemon") self.looping_call_manager.start(Checker.INTERNET_CONNECTION, 3600) self.looping_call_manager.start(Checker.CONNECTION_STATUS, 30) self.exchange_rate_manager.start() yield self._initial_setup() yield threads.deferToThread(self._setup_data_directory) yield self._check_db_migration() yield self._get_session() yield self._get_analytics() yield add_lbry_file_to_sd_identifier(self.sd_identifier) yield self._setup_stream_identifier() yield self._setup_lbry_file_manager() yield self._setup_query_handlers() yield self._setup_server() log.info("Starting balance: " + str(self.session.wallet.get_balance())) yield _announce_startup() def _get_platform(self): if self.platform is None: self.platform = system_info.get_platform() return self.platform def _initial_setup(self): def _log_platform(): log.info("Platform: %s", json.dumps(self._get_platform())) return defer.succeed(None) d = _log_platform() return d def _check_network_connection(self): self.connected_to_internet = utils.check_connection() def _check_lbrynet_connection(self): def _log_success(): log.info("lbrynet connectivity test passed") def _log_failure(): log.info("lbrynet connectivity test failed") wonderfullife_sh = ("6f3af0fa3924be98a54766aa2715d22c6c1509c3f7fa32566df4899" "a41f3530a9f97b2ecb817fa1dcbf1b30553aefaa7") d = download_sd_blob(self.session, wonderfullife_sh, self.session.base_payment_rate_manager) d.addCallbacks(lambda _: _log_success, lambda _: _log_failure) def _update_connection_status(self): self.connection_status_code = CONNECTION_STATUS_CONNECTED if not self.connected_to_internet: self.connection_status_code = CONNECTION_STATUS_NETWORK def _start_server(self): if self.peer_port is not None: server_factory = ServerProtocolFactory(self.session.rate_limiter, self.query_handlers, self.session.peer_manager) try: log.info("Daemon bound to port: %d", self.peer_port) self.lbry_server_port = reactor.listenTCP(self.peer_port, server_factory) except error.CannotListenError as e: import traceback log.error("Couldn't bind to port %d. Visit lbry.io/faq/how-to-change-port for" " more details.", self.peer_port) log.error("%s", traceback.format_exc()) raise ValueError("%s lbrynet may already be running on your computer.", str(e)) return defer.succeed(True) def _start_reflector(self): if self.run_reflector_server: log.info("Starting reflector server") if self.reflector_port is not None: reflector_factory = reflector_server_factory( self.session.peer_manager, self.session.blob_manager, self.stream_info_manager ) try: self.reflector_server_port = reactor.listenTCP(self.reflector_port, reflector_factory) log.info('Started reflector on port %s', self.reflector_port) except error.CannotListenError as e: log.exception("Couldn't bind reflector to port %d", self.reflector_port) raise ValueError( "{} lbrynet may already be running on your computer.".format(e)) return defer.succeed(True) def _stop_reflector(self): if self.run_reflector_server: log.info("Stopping reflector server") try: if self.reflector_server_port is not None: self.reflector_server_port, p = None, self.reflector_server_port return defer.maybeDeferred(p.stopListening) except AttributeError: return defer.succeed(True) return defer.succeed(True) def _stop_file_manager(self): if self.lbry_file_manager: self.lbry_file_manager.stop() return defer.succeed(True) def _stop_server(self): try: if self.lbry_server_port is not None: self.lbry_server_port, old_port = None, self.lbry_server_port log.info('Stop listening on port %s', old_port.port) return defer.maybeDeferred(old_port.stopListening) else: return defer.succeed(True) except AttributeError: return defer.succeed(True) def _setup_server(self): self.startup_status = STARTUP_STAGES[4] d = self._start_server() d.addCallback(lambda _: self._start_reflector()) return d def _setup_query_handlers(self): handlers = [ BlobRequestHandlerFactory( self.session.blob_manager, self.session.wallet, self.session.payment_rate_manager, self.analytics_manager ), self.session.wallet.get_wallet_info_query_handler_factory(), ] return self._add_query_handlers(handlers) def _add_query_handlers(self, query_handlers): for handler in query_handlers: query_id = handler.get_primary_query_identifier() self.query_handlers[query_id] = handler return defer.succeed(None) def _clean_up_temp_files(self): for path in self.uploaded_temp_files: try: log.debug('Removing tmp file: %s', path) os.remove(path) except OSError: pass @staticmethod def _already_shutting_down(sig_num, frame): log.info("Already shutting down") def _stop_streams(self): """stop pending GetStream downloads""" for claim_id, stream in self.streams.iteritems(): stream.cancel(reason="daemon shutdown") def _shutdown(self): # ignore INT/TERM signals once shutdown has started signal.signal(signal.SIGINT, self._already_shutting_down) signal.signal(signal.SIGTERM, self._already_shutting_down) log.info("Closing lbrynet session") log.info("Status at time of shutdown: " + self.startup_status[0]) self._stop_streams() self.looping_call_manager.shutdown() if self.analytics_manager: self.analytics_manager.shutdown() self._clean_up_temp_files() d = self._stop_server() d.addErrback(log.fail(), 'Failure while shutting down') d.addCallback(lambda _: self._stop_reflector()) d.addErrback(log.fail(), 'Failure while shutting down') d.addCallback(lambda _: self._stop_file_manager()) d.addErrback(log.fail(), 'Failure while shutting down') if self.session is not None: d.addCallback(lambda _: self.session.shut_down()) d.addErrback(log.fail(), 'Failure while shutting down') return d def _update_settings(self, settings): setting_types = { 'download_directory': str, 'data_rate': float, 'download_timeout': int, 'peer_port': int, 'max_key_fee': dict, 'use_upnp': bool, 'run_reflector_server': bool, 'cache_time': int, 'reflect_uploads': bool, 'share_usage_data': bool, 'disable_max_key_fee': bool, 'peer_search_timeout': int, 'sd_download_timeout': int, } for key, setting_type in setting_types.iteritems(): if key in settings: if isinstance(settings[key], setting_type): conf.settings.update({key: settings[key]}, data_types=(conf.TYPE_RUNTIME, conf.TYPE_PERSISTED)) elif setting_type is dict and isinstance(settings[key], (unicode, str)): decoded = json.loads(str(settings[key])) conf.settings.update({key: decoded}, data_types=(conf.TYPE_RUNTIME, conf.TYPE_PERSISTED)) else: try: converted = setting_type(settings[key]) conf.settings.update({key: converted}, data_types=(conf.TYPE_RUNTIME, conf.TYPE_PERSISTED)) except Exception as err: log.warning(err.message) log.warning("error converting setting '%s' to type %s from type %s", key, setting_type, str(type(settings[key]))) conf.settings.save_conf_file_settings() self.data_rate = conf.settings['data_rate'] self.max_key_fee = conf.settings['max_key_fee'] self.disable_max_key_fee = conf.settings['disable_max_key_fee'] self.download_directory = conf.settings['download_directory'] self.download_timeout = conf.settings['download_timeout'] return defer.succeed(True) def _write_db_revision_file(self, version_num): with open(self.db_revision_file, mode='w') as db_revision: db_revision.write(str(version_num)) def _setup_data_directory(self): old_revision = 1 self.startup_status = STARTUP_STAGES[1] log.info("Loading databases") if not os.path.exists(self.download_directory): os.mkdir(self.download_directory) if not os.path.exists(self.db_dir): os.mkdir(self.db_dir) self._write_db_revision_file(self.current_db_revision) log.debug("Created the db revision file: %s", self.db_revision_file) if not os.path.exists(self.blobfile_dir): os.mkdir(self.blobfile_dir) log.debug("Created the blobfile directory: %s", str(self.blobfile_dir)) if not os.path.exists(self.db_revision_file): log.warning("db_revision file not found. Creating it") self._write_db_revision_file(old_revision) def _check_db_migration(self): old_revision = 1 if os.path.exists(self.db_revision_file): old_revision = int(open(self.db_revision_file).read().strip()) if old_revision > self.current_db_revision: raise Exception('This version of lbrynet is not compatible with the database\n' 'Your database is revision %i, expected %i' % (old_revision, self.current_db_revision)) def update_version_file_and_print_success(): self._write_db_revision_file(self.current_db_revision) log.info("Finished upgrading the databases.") if old_revision < self.current_db_revision: from lbrynet.db_migrator import dbmigrator log.info("Upgrading your databases") d = threads.deferToThread( dbmigrator.migrate_db, self.db_dir, old_revision, self.current_db_revision) d.addCallback(lambda _: update_version_file_and_print_success()) return d return defer.succeed(True) @defer.inlineCallbacks def _setup_lbry_file_manager(self): log.info('Starting to setup up file manager') self.startup_status = STARTUP_STAGES[3] self.stream_info_manager = DBEncryptedFileMetadataManager(self.db_dir) yield self.stream_info_manager.setup() self.lbry_file_manager = EncryptedFileManager( self.session, self.stream_info_manager, self.sd_identifier, download_directory=self.download_directory ) yield self.lbry_file_manager.setup() log.info('Done setting up file manager') def _get_analytics(self): if not self.analytics_manager.is_started: self.analytics_manager.start() self.analytics_manager.register_repeating_metric( analytics.BLOB_BYTES_AVAILABLE, AlwaysSend(calculate_available_blob_size, self.session.blob_manager), frequency=300 ) def _get_session(self): def get_wallet(): if self.wallet_type == LBRYCRD_WALLET: raise ValueError('LBRYcrd Wallet is no longer supported') elif self.wallet_type == LBRYUM_WALLET: log.info("Using lbryum wallet") config = {'auto_connect': True} if conf.settings['lbryum_wallet_dir']: config['lbryum_path'] = conf.settings['lbryum_wallet_dir'] storage = SqliteStorage(self.db_dir) wallet = LBRYumWallet(storage, config) return defer.succeed(wallet) elif self.wallet_type == PTC_WALLET: log.info("Using PTC wallet") from lbrynet.core.PTCWallet import PTCWallet return defer.succeed(PTCWallet(self.db_dir)) else: raise ValueError('Wallet Type {} is not valid'.format(self.wallet_type)) d = get_wallet() def create_session(wallet): self.session = Session( conf.settings['data_rate'], db_dir=self.db_dir, node_id=self.node_id, blob_dir=self.blobfile_dir, dht_node_port=self.dht_node_port, known_dht_nodes=conf.settings['known_dht_nodes'], peer_port=self.peer_port, use_upnp=self.use_upnp, wallet=wallet, is_generous=conf.settings['is_generous_host'], external_ip=self.platform['ip'] ) self.startup_status = STARTUP_STAGES[2] d.addCallback(create_session) d.addCallback(lambda _: self.session.setup()) return d def _setup_stream_identifier(self): file_saver_factory = EncryptedFileSaverFactory( self.session.peer_finder, self.session.rate_limiter, self.session.blob_manager, self.stream_info_manager, self.session.wallet, self.download_directory ) self.sd_identifier.add_stream_downloader_factory(EncryptedFileStreamType, file_saver_factory) return defer.succeed(None) def _download_blob(self, blob_hash, rate_manager=None, timeout=None): """ Download a blob :param blob_hash (str): blob hash :param rate_manager (PaymentRateManager), optional: the payment rate manager to use, defaults to session.payment_rate_manager :param timeout (int): blob timeout :return: BlobFile """ if not blob_hash: raise Exception("Nothing to download") rate_manager = rate_manager or self.session.payment_rate_manager timeout = timeout or 30 return download_sd_blob(self.session, blob_hash, rate_manager, timeout) @defer.inlineCallbacks def _get_stream_analytics_report(self, claim_dict): sd_hash = claim_dict.source_hash try: stream_hash = yield self.stream_info_manager.get_stream_hash_for_sd_hash(sd_hash) except Exception: stream_hash = None report = { "sd_hash": sd_hash, "stream_hash": stream_hash, } blobs = {} try: sd_host = yield self.session.blob_manager.get_host_downloaded_from(sd_hash) except Exception: sd_host = None report["sd_blob"] = sd_host if stream_hash: blob_infos = yield self.stream_info_manager.get_blobs_for_stream(stream_hash) report["known_blobs"] = len(blob_infos) else: blob_infos = [] report["known_blobs"] = 0 # for blob_hash, blob_num, iv, length in blob_infos: # try: # host = yield self.session.blob_manager.get_host_downloaded_from(blob_hash) # except Exception: # host = None # if host: # blobs[blob_num] = host # report["blobs"] = json.dumps(blobs) defer.returnValue(report) @defer.inlineCallbacks def _download_name(self, name, claim_dict, claim_id, timeout=None, file_name=None): """ Add a lbry file to the file manager, start the download, and return the new lbry file. If it already exists in the file manager, return the existing lbry file """ @defer.inlineCallbacks def _download_finished(download_id, name, claim_dict): report = yield self._get_stream_analytics_report(claim_dict) self.analytics_manager.send_download_finished(download_id, name, report, claim_dict) @defer.inlineCallbacks def _download_failed(error, download_id, name, claim_dict): report = yield self._get_stream_analytics_report(claim_dict) self.analytics_manager.send_download_errored(error, download_id, name, claim_dict, report) if claim_id in self.streams: downloader = self.streams[claim_id] result = yield downloader.finished_deferred defer.returnValue(result) else: download_id = utils.random_string() self.analytics_manager.send_download_started(download_id, name, claim_dict) self.streams[claim_id] = GetStream(self.sd_identifier, self.session, self.exchange_rate_manager, self.max_key_fee, self.disable_max_key_fee, conf.settings['data_rate'], timeout, file_name) try: lbry_file, finished_deferred = yield self.streams[claim_id].start(claim_dict, name) finished_deferred.addCallbacks(lambda _: _download_finished(download_id, name, claim_dict), lambda e: _download_failed(e, download_id, name, claim_dict)) result = yield self._get_lbry_file_dict(lbry_file, full_status=True) except Exception as err: yield _download_failed(err, download_id, name, claim_dict) if isinstance(err, (DownloadDataTimeout, DownloadSDTimeout)): log.warning('Failed to get %s (%s)', name, err) else: log.error('Failed to get %s (%s)', name, err) if self.streams[claim_id].downloader: yield self.streams[claim_id].downloader.stop(err) result = {'error': err.message} finally: del self.streams[claim_id] defer.returnValue(result) @defer.inlineCallbacks def _publish_stream(self, name, bid, claim_dict, file_path=None, certificate_id=None, claim_address=None, change_address=None): publisher = Publisher(self.session, self.lbry_file_manager, self.session.wallet, certificate_id) parse_lbry_uri(name) if bid <= 0.0: raise Exception("Invalid bid") if not file_path: claim_out = yield publisher.publish_stream(name, bid, claim_dict, claim_address, change_address) else: claim_out = yield publisher.create_and_publish_stream(name, bid, claim_dict, file_path, claim_address, change_address) if conf.settings['reflect_uploads']: d = reupload.reflect_stream(publisher.lbry_file) d.addCallbacks(lambda _: log.info("Reflected new publication to lbry://%s", name), log.exception) self.analytics_manager.send_claim_action('publish') log.info("Success! Published to lbry://%s txid: %s nout: %d", name, claim_out['txid'], claim_out['nout']) defer.returnValue(claim_out) def _get_long_count_timestamp(self): dt = utils.utcnow() - utils.datetime_obj(year=2012, month=12, day=21) return int(dt.total_seconds()) @defer.inlineCallbacks def _resolve_name(self, name, force_refresh=False): """Resolves a name. Checks the cache first before going out to the blockchain. Args: name: the lbry:// to resolve force_refresh: if True, always go out to the blockchain to resolve. """ parsed = parse_lbry_uri(name) resolution = yield self.session.wallet.resolve(parsed.name, check_cache=not force_refresh) if parsed.name in resolution: result = resolution[parsed.name] defer.returnValue(result) def _get_or_download_sd_blob(self, blob, sd_hash): if blob: return self.session.blob_manager.get_blob(blob[0]) def _check_est(downloader): if downloader.result is not None: downloader.cancel() d = defer.succeed(None) reactor.callLater(self.search_timeout, _check_est, d) d.addCallback( lambda _: download_sd_blob( self.session, sd_hash, self.session.payment_rate_manager)) return d def get_or_download_sd_blob(self, sd_hash): """Return previously downloaded sd blob if already in the blob manager, otherwise download and return it """ d = self.session.blob_manager.completed_blobs([sd_hash]) d.addCallback(self._get_or_download_sd_blob, sd_hash) return d def get_size_from_sd_blob(self, sd_blob): """ Get total stream size in bytes from a sd blob """ d = self.sd_identifier.get_metadata_for_sd_blob(sd_blob) d.addCallback(lambda metadata: metadata.validator.info_to_show()) d.addCallback(lambda info: int(dict(info)['stream_size'])) return d def _get_est_cost_from_stream_size(self, size): """ Calculate estimated LBC cost for a stream given its size in bytes """ if self.session.payment_rate_manager.generous: return 0.0 return size / (10 ** 6) * conf.settings['data_rate'] @defer.inlineCallbacks def get_est_cost_using_known_size(self, uri, size): """ Calculate estimated LBC cost for a stream given its size in bytes """ cost = self._get_est_cost_from_stream_size(size) resolved = yield self.session.wallet.resolve(uri) if uri in resolved and 'claim' in resolved[uri]: claim = ClaimDict.load_dict(resolved[uri]['claim']['value']) final_fee = self._add_key_fee_to_est_data_cost(claim.source_fee, cost) result = yield self._render_response(final_fee) defer.returnValue(result) else: defer.returnValue(None) def get_est_cost_from_sd_hash(self, sd_hash): """ Get estimated cost from a sd hash """ d = self.get_or_download_sd_blob(sd_hash) d.addCallback(self.get_size_from_sd_blob) d.addCallback(self._get_est_cost_from_stream_size) return d def _get_est_cost_from_metadata(self, metadata, name): d = self.get_est_cost_from_sd_hash(metadata.source_hash) def _handle_err(err): if isinstance(err, Failure): log.warning( "Timeout getting blob for cost est for lbry://%s, using only key fee", name) return 0.0 raise err d.addErrback(_handle_err) d.addCallback(lambda data_cost: self._add_key_fee_to_est_data_cost(metadata.source_fee, data_cost)) return d def _add_key_fee_to_est_data_cost(self, fee, data_cost): fee_amount = 0.0 if not fee else self.exchange_rate_manager.convert_currency(fee.currency, "LBC", fee.amount) return data_cost + fee_amount @defer.inlineCallbacks def get_est_cost_from_uri(self, uri): """ Resolve a name and return the estimated stream cost """ resolved = yield self.session.wallet.resolve(uri) if resolved: claim_response = resolved[uri] else: claim_response = None result = None if claim_response and 'claim' in claim_response: if 'value' in claim_response['claim'] and claim_response['claim']['value'] is not None: claim_value = ClaimDict.load_dict(claim_response['claim']['value']) cost = yield self._get_est_cost_from_metadata(claim_value, uri) result = round(cost, 5) else: log.warning("Failed to estimate cost for %s", uri) defer.returnValue(result) def get_est_cost(self, uri, size=None): """Get a cost estimate for a lbry stream, if size is not provided the sd blob will be downloaded to determine the stream size """ if size is not None: return self.get_est_cost_using_known_size(uri, size) return self.get_est_cost_from_uri(uri) @defer.inlineCallbacks def _get_lbry_file_dict(self, lbry_file, full_status=False): key = binascii.b2a_hex(lbry_file.key) if lbry_file.key else None full_path = os.path.join(lbry_file.download_directory, lbry_file.file_name) mime_type = mimetypes.guess_type(full_path)[0] if os.path.isfile(full_path): with open(full_path) as written_file: written_file.seek(0, os.SEEK_END) written_bytes = written_file.tell() else: written_bytes = 0 if full_status: size = yield lbry_file.get_total_bytes() file_status = yield lbry_file.status() message = STREAM_STAGES[2][1] % (file_status.name, file_status.num_completed, file_status.num_known, file_status.running_status) else: size = None message = None claim = yield self.session.wallet.get_claim_by_claim_id(lbry_file.claim_id, check_expire=False) if claim and 'value' in claim: metadata = claim['value'] else: metadata = None if claim and 'channel_name' in claim: channel_name = claim['channel_name'] else: channel_name = None if lbry_file.txid and lbry_file.nout is not None: outpoint = repr(ClaimOutpoint(lbry_file.txid, lbry_file.nout)) else: outpoint = None if claim and 'has_signature' in claim: has_signature = claim['has_signature'] else: has_signature = None if claim and 'signature_is_valid' in claim: signature_is_valid = claim['signature_is_valid'] else: signature_is_valid = None result = { 'completed': lbry_file.completed, 'file_name': lbry_file.file_name, 'download_directory': lbry_file.download_directory, 'points_paid': lbry_file.points_paid, 'stopped': lbry_file.stopped, 'stream_hash': lbry_file.stream_hash, 'stream_name': lbry_file.stream_name, 'suggested_file_name': lbry_file.suggested_file_name, 'sd_hash': lbry_file.sd_hash, 'name': lbry_file.name, 'outpoint': outpoint, 'claim_id': lbry_file.claim_id, 'download_path': full_path, 'mime_type': mime_type, 'key': key, 'total_bytes': size, 'written_bytes': written_bytes, 'message': message, 'metadata': metadata } if channel_name is not None: result['channel_name'] = channel_name if has_signature is not None: result['has_signature'] = has_signature if signature_is_valid is not None: result['signature_is_valid'] = signature_is_valid defer.returnValue(result) @defer.inlineCallbacks def _get_lbry_file(self, search_by, val, return_json=False, full_status=False): lbry_file = None if search_by in FileID: for l_f in self.lbry_file_manager.lbry_files: if l_f.__dict__.get(search_by) == val: lbry_file = l_f break else: raise NoValidSearch('{} is not a valid search operation'.format(search_by)) if return_json and lbry_file: lbry_file = yield self._get_lbry_file_dict(lbry_file, full_status=full_status) defer.returnValue(lbry_file) @defer.inlineCallbacks def _get_lbry_files(self, return_json=False, full_status=False, **kwargs): lbry_files = list(self.lbry_file_manager.lbry_files) if kwargs: for search_type, value in iter_lbry_file_search_values(kwargs): lbry_files = [l_f for l_f in lbry_files if l_f.__dict__[search_type] == value] if return_json: file_dicts = [] for lbry_file in lbry_files: lbry_file_dict = yield self._get_lbry_file_dict(lbry_file, full_status=full_status) file_dicts.append(lbry_file_dict) lbry_files = file_dicts log.debug("Collected %i lbry files", len(lbry_files)) defer.returnValue(lbry_files) # TODO: do this and get_blobs_for_sd_hash in the stream info manager def get_blobs_for_stream_hash(self, stream_hash): def _iter_blobs(blob_hashes): for blob_hash, blob_num, blob_iv, blob_length in blob_hashes: if blob_hash: yield self.session.blob_manager.get_blob(blob_hash, length=blob_length) def _get_blobs(blob_hashes): dl = defer.DeferredList(list(_iter_blobs(blob_hashes)), consumeErrors=True) dl.addCallback(lambda blobs: [blob[1] for blob in blobs if blob[0]]) return dl d = self.stream_info_manager.get_blobs_for_stream(stream_hash) d.addCallback(_get_blobs) return d def get_blobs_for_sd_hash(self, sd_hash): d = self.stream_info_manager.get_stream_hash_for_sd_hash(sd_hash) d.addCallback(self.get_blobs_for_stream_hash) return d ############################################################################ # # # JSON-RPC API methods start here # # # ############################################################################ @defer.inlineCallbacks @AuthJSONRPCServer.flags(session_status="-s", dht_status="-d") def jsonrpc_status(self, session_status=False, dht_status=False): """ Get daemon status Usage: status [-s] [-d] Options: -s : include session status in results -d : include dht network and peer status Returns: (dict) lbrynet-daemon status { 'lbry_id': lbry peer id, base58 'installation_id': installation id, base58 'is_running': bool 'is_first_run': bool 'startup_status': { 'code': status code 'message': status message }, 'connection_status': { 'code': connection status code 'message': connection status message }, 'blockchain_status': { 'blocks': local blockchain height, 'blocks_behind': remote_height - local_height, 'best_blockhash': block hash of most recent block, }, If given the session status option: 'session_status': { 'managed_blobs': count of blobs in the blob manager, 'managed_streams': count of streams in the file manager 'announce_queue_size': number of blobs currently queued to be announced 'should_announce_blobs': number of blobs that should be announced } If given the dht status option: 'dht_status': { 'kbps_received': current kbps receiving, 'kbps_sent': current kdps being sent, 'total_bytes_sent': total bytes sent 'total_bytes_received': total bytes received 'queries_received': number of queries received per second 'queries_sent': number of queries sent per second 'recent_contacts': count of recently contacted peers 'unique_contacts': count of unique peers } } """ # on startup, the wallet or network won't be available but we still need this call to work has_wallet = self.session and self.session.wallet and self.session.wallet.network local_height = self.session.wallet.network.get_local_height() if has_wallet else 0 remote_height = self.session.wallet.network.get_server_height() if has_wallet else 0 best_hash = (yield self.session.wallet.get_best_blockhash()) if has_wallet else None response = { 'lbry_id': base58.b58encode(self.node_id), 'installation_id': conf.settings.installation_id, 'is_running': self.announced_startup, 'is_first_run': self.session.wallet.is_first_run if has_wallet else None, 'startup_status': { 'code': self.startup_status[0], 'message': self.startup_status[1], }, 'connection_status': { 'code': self.connection_status_code, 'message': ( CONNECTION_MESSAGES[self.connection_status_code] if self.connection_status_code is not None else '' ), }, 'blocks_behind': remote_height - local_height, # deprecated. remove from UI, then here 'blockchain_status': { 'blocks': local_height, 'blocks_behind': remote_height - local_height, 'best_blockhash': best_hash, } } if session_status: blobs = yield self.session.blob_manager.get_all_verified_blobs() announce_queue_size = self.session.hash_announcer.hash_queue_size() should_announce_blobs = yield self.session.blob_manager.count_should_announce_blobs() response['session_status'] = { 'managed_blobs': len(blobs), 'managed_streams': len(self.lbry_file_manager.lbry_files), 'announce_queue_size': announce_queue_size, 'should_announce_blobs': should_announce_blobs, } if dht_status: response['dht_status'] = self.session.dht_node.get_bandwidth_stats() defer.returnValue(response) def jsonrpc_version(self): """ Get lbry version information Usage: version Returns: (dict) Dictionary of lbry version information { 'build': (str) build type (e.g. "dev", "rc", "release"), 'ip': (str) remote ip, if available, 'lbrynet_version': (str) lbrynet_version, 'lbryum_version': (str) lbryum_version, 'lbryschema_version': (str) lbryschema_version, 'os_release': (str) os release string 'os_system': (str) os name 'platform': (str) platform string 'processor': (str) processor type, 'python_version': (str) python version, } """ platform_info = self._get_platform() log.info("Get version info: " + json.dumps(platform_info)) return self._render_response(platform_info) def jsonrpc_report_bug(self, message=None): """ Report a bug to slack Usage: report_bug ( | --message=) Returns: (bool) true if successful """ platform_name = self._get_platform()['platform'] report_bug_to_slack( message, conf.settings.installation_id, platform_name, get_lbrynet_version() ) return self._render_response(True) def jsonrpc_settings_get(self): """ Get daemon settings Usage: settings_get Returns: (dict) Dictionary of daemon settings See ADJUSTABLE_SETTINGS in lbrynet/conf.py for full list of settings """ return self._render_response(conf.settings.get_adjustable_settings_dict()) @AuthJSONRPCServer.auth_required @defer.inlineCallbacks def jsonrpc_settings_set(self, **kwargs): """ Set daemon settings Usage: settings_set [ | --download_directory=] [ | --data_rate=] [ | --download_timeout=] [ | --peer_port=] [ | --max_key_fee=] [ | --disable_max_key_fee=] [ | --use_upnp=] [ | --run_reflector_server=] [ | --cache_time=] [ | --reflect_uploads=] [ | --share_usage_data=] [ | --peer_search_timeout=] [ | --sd_download_timeout=] Options: , --download_directory= : (str) , --data_rate= : (float), 0.0001 , --download_timeout= : (int), 180 , --peer_port= : (int), 3333 , --max_key_fee= : (dict) maximum key fee for downloads, in the format: { "currency": , "amount": }. In the CLI, it must be an escaped JSON string Supported currency symbols: LBC BTC USD , --disable_max_key_fee= : (bool), False , --use_upnp= : (bool), True , --run_reflector_server= : (bool), False , --cache_time= : (int), 150 , --reflect_uploads= : (bool), True , --share_usage_data= : (bool), True , --peer_search_timeout= : (int), 3 , --sd_download_timeout= : (int), 3 Returns: (dict) Updated dictionary of daemon settings """ yield self._update_settings(kwargs) defer.returnValue(conf.settings.get_adjustable_settings_dict()) def jsonrpc_help(self, command=None): """ Return a useful message for an API command Usage: help [ | --command=] Options: , --command= : command to retrieve documentation for """ if command is None: return self._render_response({ 'about': 'This is the LBRY JSON-RPC API', 'command_help': 'Pass a `command` parameter to this method to see ' + 'help for that command (e.g. `help command=resolve_name`)', 'command_list': 'Get a full list of commands using the `commands` method', 'more_info': 'Visit https://lbry.io/api for more info', }) fn = self.callable_methods.get(command) if fn is None: raise Exception( "No help available for '{}'. It is not a valid command.".format(command) ) return self._render_response({ 'help': textwrap.dedent(fn.__doc__ or '') }) def jsonrpc_commands(self): """ Return a list of available commands Usage: commands Returns: (list) list of available commands """ return self._render_response(sorted([command for command in self.callable_methods.keys()])) @AuthJSONRPCServer.flags(include_unconfirmed='-u') def jsonrpc_wallet_balance(self, address=None, include_unconfirmed=False): """ Return the balance of the wallet Usage: wallet_balance [
| --address=
] [-u] Options:
: If provided only the balance for this address will be given -u : Include unconfirmed Returns: (float) amount of lbry credits in wallet """ if address is None: return self._render_response(float(self.session.wallet.get_balance())) else: return self._render_response(float( self.session.wallet.get_address_balance(address, include_unconfirmed))) @defer.inlineCallbacks def jsonrpc_daemon_stop(self): """ Stop lbrynet-daemon Usage: daemon_stop Returns: (string) Shutdown message """ log.info("Shutting down lbrynet daemon") response = yield self._render_response("Shutting down") reactor.callLater(0.1, reactor.fireSystemEvent, "shutdown") defer.returnValue(response) @defer.inlineCallbacks @AuthJSONRPCServer.flags(full_status='-f') def jsonrpc_file_list(self, **kwargs): """ List files limited by optional filters Usage: file_list [--sd_hash=] [--file_name=] [--stream_hash=] [--claim_id=] [--outpoint=] [--rowid=] [--name=] [-f] Options: --sd_hash= : get file with matching sd hash --file_name= : get file with matching file name in the downloads folder --stream_hash= : get file with matching stream hash --claim_id= : get file with matching claim id --outpoint= : get file with matching claim outpoint --rowid= : get file with matching row id --name= : get file with matching associated name claim -f : full status, populate the 'message' and 'size' fields Returns: (list) List of files [ { 'completed': (bool) true if download is completed, 'file_name': (str) name of file, 'download_directory': (str) download directory, 'points_paid': (float) credit paid to download file, 'stopped': (bool) true if download is stopped, 'stream_hash': (str) stream hash of file, 'stream_name': (str) stream name , 'suggested_file_name': (str) suggested file name, 'sd_hash': (str) sd hash of file, 'name': (str) name claim attached to file 'outpoint': (str) claim outpoint attached to file 'claim_id': (str) claim ID attached to file, 'download_path': (str) download path of file, 'mime_type': (str) mime type of file, 'key': (str) key attached to file, 'total_bytes': (int) file size in bytes, None if full_status is false 'written_bytes': (int) written size in bytes 'message': (str), None if full_status is false 'metadata': (dict) Metadata dictionary }, ] """ result = yield self._get_lbry_files(return_json=True, **kwargs) response = yield self._render_response(result) defer.returnValue(response) @defer.inlineCallbacks @AuthJSONRPCServer.flags(force='-f') def jsonrpc_resolve_name(self, name, force=False): """ Resolve stream info from a LBRY name Usage: resolve_name [-f] Options: -f : force refresh and do not check cache Returns: (dict) Metadata dictionary from name claim, None if the name is not resolvable """ try: metadata = yield self._resolve_name(name, force_refresh=force) except UnknownNameError: log.info('Name %s is not known', name) defer.returnValue(None) else: defer.returnValue(metadata) @defer.inlineCallbacks def jsonrpc_claim_show(self, txid=None, nout=None, claim_id=None): """ Resolve claim info from txid/nout or with claim ID Usage: claim_show [ | --txid=] [ | --nout=] [ | --claim_id=] Options: , --txid= : look for claim with this txid, nout must also be specified , --nout= : look for claim with this nout, txid must also be specified , --claim_id= : look for claim with this claim id Returns: (dict) Dictionary containing claim info as below, { 'txid': (str) txid of claim 'nout': (int) nout of claim 'amount': (float) amount of claim 'value': (str) value of claim 'height' : (int) height of claim takeover 'claim_id': (str) claim ID of claim 'supports': (list) list of supports associated with claim } if claim cannot be resolved, dictionary as below will be returned { 'error': (str) reason for error } """ if claim_id is not None and txid is None and nout is None: claim_results = yield self.session.wallet.get_claim_by_claim_id(claim_id) elif txid is not None and nout is not None and claim_id is None: outpoint = ClaimOutpoint(txid, nout) claim_results = yield self.session.wallet.get_claim_by_outpoint(outpoint) else: raise Exception("Must specify either txid/nout, or claim_id") response = yield self._render_response(claim_results) defer.returnValue(response) @AuthJSONRPCServer.auth_required @defer.inlineCallbacks @AuthJSONRPCServer.flags(force='-f') def jsonrpc_resolve(self, force=False, uri=None, uris=[]): """ Resolve given LBRY URIs Usage: resolve [-f] ( | --uri=) [...] Options: -f : force refresh and ignore cache Returns: Dictionary of results, keyed by uri '': { If a resolution error occurs: 'error': Error message If the uri resolves to a channel or a claim in a channel: 'certificate': { 'address': (str) claim address, 'amount': (float) claim amount, 'effective_amount': (float) claim amount including supports, 'claim_id': (str) claim id, 'claim_sequence': (int) claim sequence number, 'decoded_claim': (bool) whether or not the claim value was decoded, 'height': (int) claim height, 'depth': (int) claim depth, 'has_signature': (bool) included if decoded_claim 'name': (str) claim name, 'supports: (list) list of supports [{'txid': (str) txid, 'nout': (int) nout, 'amount': (float) amount}], 'txid': (str) claim txid, 'nout': (str) claim nout, 'signature_is_valid': (bool), included if has_signature, 'value': ClaimDict if decoded, otherwise hex string } If the uri resolves to a channel: 'claims_in_channel': (int) number of claims in the channel, If the uri resolves to a claim: 'claim': { 'address': (str) claim address, 'amount': (float) claim amount, 'effective_amount': (float) claim amount including supports, 'claim_id': (str) claim id, 'claim_sequence': (int) claim sequence number, 'decoded_claim': (bool) whether or not the claim value was decoded, 'height': (int) claim height, 'depth': (int) claim depth, 'has_signature': (bool) included if decoded_claim 'name': (str) claim name, 'channel_name': (str) channel name if claim is in a channel 'supports: (list) list of supports [{'txid': (str) txid, 'nout': (int) nout, 'amount': (float) amount}] 'txid': (str) claim txid, 'nout': (str) claim nout, 'signature_is_valid': (bool), included if has_signature, 'value': ClaimDict if decoded, otherwise hex string } } """ uris = tuple(uris) if uri is not None: uris += (uri,) results = {} valid_uris = tuple() for u in uris: try: parse_lbry_uri(u) valid_uris += (u, ) except URIParseError: results[u] = {"error": "%s is not a valid uri" % u} resolved = yield self.session.wallet.resolve(*valid_uris, check_cache=not force) for resolved_uri in resolved: results[resolved_uri] = resolved[resolved_uri] response = yield self._render_response(results) defer.returnValue(response) @AuthJSONRPCServer.auth_required @defer.inlineCallbacks def jsonrpc_get(self, uri, file_name=None, timeout=None): """ Download stream from a LBRY name. Usage: get [ | --file_name=] [ | --timeout=] Options: : specified name for the downloaded file : download timeout in number of seconds : path to directory where file will be saved Returns: (dict) Dictionary containing information about the stream { 'completed': (bool) true if download is completed, 'file_name': (str) name of file, 'download_directory': (str) download directory, 'points_paid': (float) credit paid to download file, 'stopped': (bool) true if download is stopped, 'stream_hash': (str) stream hash of file, 'stream_name': (str) stream name, 'suggested_file_name': (str) suggested file name, 'sd_hash': (str) sd hash of file, 'name': (str) name claim attached to file 'outpoint': (str) claim outpoint attached to file 'claim_id': (str) claim ID attached to file, 'download_path': (str) download path of file, 'mime_type': (str) mime type of file, 'key': (str) key attached to file, 'total_bytes': (int) file size in bytes, None if full_status is false 'written_bytes': (int) written size in bytes 'message': (str), None if full_status is false 'metadata': (dict) Metadata dictionary } """ timeout = timeout if timeout is not None else self.download_timeout resolved_result = yield self.session.wallet.resolve(uri) if resolved_result and uri in resolved_result: resolved = resolved_result[uri] else: resolved = None if not resolved or 'value' not in resolved: if 'claim' not in resolved: raise Exception( "Failed to resolve stream at lbry://{}".format(uri.replace("lbry://", ""))) else: resolved = resolved['claim'] name = resolved['name'] claim_id = resolved['claim_id'] claim_dict = ClaimDict.load_dict(resolved['value']) if claim_id in self.streams: log.info("Already waiting on lbry://%s to start downloading", name) yield self.streams[claim_id].data_downloading_deferred lbry_file = yield self._get_lbry_file(FileID.CLAIM_ID, claim_id, return_json=False) if lbry_file: if not os.path.isfile(os.path.join(lbry_file.download_directory, lbry_file.file_name)): log.info("Already have lbry file but missing file in %s, rebuilding it", lbry_file.download_directory) yield lbry_file.start() else: log.info('Already have a file for %s', name) result = yield self._get_lbry_file_dict(lbry_file, full_status=True) else: result = yield self._download_name(name, claim_dict, claim_id, timeout=timeout, file_name=file_name) response = yield self._render_response(result) defer.returnValue(response) @AuthJSONRPCServer.auth_required @defer.inlineCallbacks def jsonrpc_file_set_status(self, status, **kwargs): """ Start or stop downloading a file Usage: file_set_status [--sd_hash=] [--file_name=] [--stream_hash=] [--claim_id=] [--outpoint=] [--rowid=] [--name=] Options: --sd_hash= : set status of file with matching sd hash --file_name= : set status of file with matching file name in the downloads folder --stream_hash= : set status of file with matching stream hash --claim_id= : set status of file with matching claim id --outpoint= : set status of file with matching claim outpoint --rowid= : set status of file with matching row id --name= : set status of file with matching associated name claim Returns: (str) Confirmation message """ if status not in ['start', 'stop']: raise Exception('Status must be "start" or "stop".') search_type, value = get_lbry_file_search_value(kwargs) lbry_file = yield self._get_lbry_file(search_type, value, return_json=False) if not lbry_file: raise Exception('Unable to find a file for {}:{}'.format(search_type, value)) if status == 'start' and lbry_file.stopped or status == 'stop' and not lbry_file.stopped: yield self.lbry_file_manager.toggle_lbry_file_running(lbry_file) msg = "Started downloading file" if status == 'start' else "Stopped downloading file" else: msg = ( "File was already being downloaded" if status == 'start' else "File was already stopped" ) response = yield self._render_response(msg) defer.returnValue(response) @AuthJSONRPCServer.auth_required @defer.inlineCallbacks @AuthJSONRPCServer.flags(delete_from_download_dir='-f', delete_all='--delete_all') def jsonrpc_file_delete(self, delete_from_download_dir=False, delete_all=False, **kwargs): """ Delete a LBRY file Usage: file_delete [-f] [--delete_all] [--sd_hash=] [--file_name=] [--stream_hash=] [--claim_id=] [--outpoint=] [--rowid=] [--name=] Options: -f, --delete_from_download_dir : delete file from download directory, instead of just deleting blobs --delete_all : if there are multiple matching files, allow the deletion of multiple files. Otherwise do not delete anything. --sd_hash= : delete by file sd hash --file_name : delete by file name in downloads folder --stream_hash= : delete by file stream hash --claim_id= : delete by file claim id --outpoint= : delete by file claim outpoint --rowid= : delete by file row id --name= : delete by associated name claim of file Returns: (bool) true if deletion was successful """ lbry_files = yield self._get_lbry_files(return_json=False, **kwargs) if len(lbry_files) > 1: if not delete_all: log.warning("There are %i files to delete, use narrower filters to select one", len(lbry_files)) response = yield self._render_response(False) defer.returnValue(response) else: log.warning("Deleting %i files", len(lbry_files)) if not lbry_files: log.warning("There is no file to delete") result = False else: for lbry_file in lbry_files: file_name, stream_hash = lbry_file.file_name, lbry_file.stream_hash if lbry_file.claim_id in self.streams: del self.streams[lbry_file.claim_id] yield self.lbry_file_manager.delete_lbry_file(lbry_file, delete_file=delete_from_download_dir) log.info("Deleted file: %s", file_name) result = True response = yield self._render_response(result) defer.returnValue(response) @defer.inlineCallbacks def jsonrpc_stream_cost_estimate(self, uri, size=None): """ Get estimated cost for a lbry stream Usage: stream_cost_estimate [ | --size=] Options: , --size= : stream size in bytes. if provided an sd blob won't be downloaded. Returns: (float) Estimated cost in lbry credits, returns None if uri is not resolveable """ cost = yield self.get_est_cost(uri, size) defer.returnValue(cost) @AuthJSONRPCServer.auth_required @AuthJSONRPCServer.queued @defer.inlineCallbacks def jsonrpc_channel_new(self, channel_name, amount): """ Generate a publisher key and create a new '@' prefixed certificate claim Usage: channel_new ( | --channel_name=) ( | --amount=) Returns: (dict) Dictionary containing result of the claim { 'tx' : (str) hex encoded transaction 'txid' : (str) txid of resulting claim 'nout' : (int) nout of the resulting claim 'fee' : (float) fee paid for the claim transaction 'claim_id' : (str) claim ID of the resulting claim } """ try: parsed = parse_lbry_uri(channel_name) if not parsed.is_channel: raise Exception("Cannot make a new channel for a non channel name") if parsed.path: raise Exception("Invalid channel uri") except (TypeError, URIParseError): raise Exception("Invalid channel name") if amount <= 0: raise Exception("Invalid amount") if amount > self.session.wallet.get_balance(): raise InsufficientFundsError() result = yield self.session.wallet.claim_new_channel(channel_name, amount) self.analytics_manager.send_new_channel() log.info("Claimed a new channel! Result: %s", result) response = yield self._render_response(result) defer.returnValue(response) @AuthJSONRPCServer.auth_required @defer.inlineCallbacks def jsonrpc_channel_list_mine(self): """ Get my channels Usage: channel_list_mine Returns: (list) ClaimDict """ result = yield self.session.wallet.channel_list() response = yield self._render_response(result) defer.returnValue(response) @AuthJSONRPCServer.auth_required @AuthJSONRPCServer.queued @defer.inlineCallbacks def jsonrpc_publish(self, name, bid, metadata=None, file_path=None, fee=None, title=None, description=None, author=None, language=None, license=None, license_url=None, thumbnail=None, preview=None, nsfw=None, sources=None, channel_name=None, channel_id=None, claim_address=None, change_address=None): """ Make a new name claim and publish associated data to lbrynet, update over existing claim if user already has a claim for name. Fields required in the final Metadata are: 'title' 'description' 'author' 'language' 'license' 'nsfw' Metadata can be set by either using the metadata argument or by setting individual arguments fee, title, description, author, language, license, license_url, thumbnail, preview, nsfw, or sources. Individual arguments will overwrite the fields specified in metadata argument. Usage: publish ( | --name=) ( | --bid=) [--metadata=] [--file_path=] [--fee=] [--title=] [--description=<description>] [--author=<author>] [--language=<language>] [--license=<license>] [--license_url=<license_url>] [--thumbnail=<thumbnail>] [--preview=<preview>] [--nsfw=<nsfw>] [--sources=<sources>] [--channel_name=<channel_name>] [--channel_id=<channel_id>] [--claim_address=<claim_address>] [--change_address=<change_address>] Options: --metadata=<metadata> : ClaimDict to associate with the claim. --file_path=<file_path> : path to file to be associated with name. If provided, a lbry stream of this file will be used in 'sources'. If no path is given but a metadata dict is provided, the source from the given metadata will be used. --fee=<fee> : Dictionary representing key fee to download content: { 'currency': currency_symbol, 'amount': float, 'address': str, optional } supported currencies: LBC, USD, BTC If an address is not provided a new one will be automatically generated. Default fee is zero. --title=<title> : title of the publication --description=<description> : description of the publication --author=<author> : author of the publication --language=<language> : language of the publication --license=<license> : publication license --license_url=<license_url> : publication license url --thumbnail=<thumbnail> : thumbnail url --preview=<preview> : preview url --nsfw=<nsfw> : title of the publication --sources=<sources> : {'lbry_sd_hash':sd_hash} specifies sd hash of file --channel_name=<channel_name> : name of the publisher channel name in the wallet --channel_id=<channel_id> : claim id of the publisher channel, does not check for channel claim being in the wallet. This allows publishing to a channel where only the certificate private key is in the wallet. --claim_address=<claim_address> : address where the claim is sent to, if not specified new address wil automatically be created Returns: (dict) Dictionary containing result of the claim { 'tx' : (str) hex encoded transaction 'txid' : (str) txid of resulting claim 'nout' : (int) nout of the resulting claim 'fee' : (float) fee paid for the claim transaction 'claim_id' : (str) claim ID of the resulting claim } """ try: parse_lbry_uri(name) except (TypeError, URIParseError): raise Exception("Invalid name given to publish") if bid <= 0.0: raise Exception("Invalid bid") if bid >= self.session.wallet.get_balance(): raise InsufficientFundsError('Insufficient funds. ' \ 'Make sure you have enough LBC to deposit') metadata = metadata or {} if fee is not None: metadata['fee'] = fee if title is not None: metadata['title'] = title if description is not None: metadata['description'] = description if author is not None: metadata['author'] = author if language is not None: metadata['language'] = language if license is not None: metadata['license'] = license if license_url is not None: metadata['licenseUrl'] = license_url if thumbnail is not None: metadata['thumbnail'] = thumbnail if preview is not None: metadata['preview'] = preview if nsfw is not None: metadata['nsfw'] = bool(nsfw) metadata['version'] = '_0_1_0' # check for original deprecated format {'currency':{'address','amount'}} # add address, version to fee if unspecified if 'fee' in metadata: if len(metadata['fee'].keys()) == 1 and isinstance(metadata['fee'].values()[0], dict): raise Exception('Old format for fee no longer supported. ' \ 'Fee must be specified as {"currency":,"address":,"amount":}') if 'amount' in metadata['fee'] and 'currency' in metadata['fee']: if not metadata['fee']['amount']: log.warning("Stripping empty fee from published metadata") del metadata['fee'] elif 'address' not in metadata['fee']: address = yield self.session.wallet.get_unused_address() metadata['fee']['address'] = address if 'fee' in metadata and 'version' not in metadata['fee']: metadata['fee']['version'] = '_0_0_1' claim_dict = { 'version': '_0_0_1', 'claimType': 'streamType', 'stream': { 'metadata': metadata, 'version': '_0_0_1' } } if sources is not None: claim_dict['stream']['source'] = sources log.info("Publish: %s", { 'name': name, 'file_path': file_path, 'bid': bid, 'claim_address': claim_address, 'change_address': change_address, 'claim_dict': claim_dict, }) if channel_id: certificate_id = channel_id elif channel_name: certificate_id = None my_certificates = yield self.session.wallet.channel_list() for certificate in my_certificates: if channel_name == certificate['name']: certificate_id = certificate['claim_id'] break if not certificate_id: raise Exception("Cannot publish using channel %s" % channel_name) else: certificate_id = None result = yield self._publish_stream(name, bid, claim_dict, file_path, certificate_id, claim_address, change_address) response = yield self._render_response(result) defer.returnValue(response) @AuthJSONRPCServer.auth_required @defer.inlineCallbacks def jsonrpc_claim_abandon(self, claim_id=None, txid=None, nout=None): """ Abandon a name and reclaim credits from the claim Usage: claim_abandon [<claim_id> | --claim_id=<claim_id>] [<txid> | --txid=<txid>] [<nout> | --nout=<nout>] Return: (dict) Dictionary containing result of the claim { txid : (str) txid of resulting transaction fee : (float) fee paid for the transaction } """ if claim_id is None and txid is None and nout is None: raise Exception('Must specify claim_id, or txid and nout') if txid is None and nout is not None: raise Exception('Must specify txid') if nout is None and txid is not None: raise Exception('Must specify nout') result = yield self.session.wallet.abandon_claim(claim_id, txid, nout) self.analytics_manager.send_claim_action('abandon') defer.returnValue(result) @AuthJSONRPCServer.auth_required @defer.inlineCallbacks def jsonrpc_claim_new_support(self, name, claim_id, amount): """ Support a name claim Usage: claim_new_support (<name> | --name=<name>) (<claim_id> | --claim_id=<claim_id>) (<amount> | --amount=<amount>) Return: (dict) Dictionary containing result of the claim { txid : (str) txid of resulting support claim nout : (int) nout of the resulting support claim fee : (float) fee paid for the transaction } """ result = yield self.session.wallet.support_claim(name, claim_id, amount) self.analytics_manager.send_claim_action('new_support') defer.returnValue(result) @AuthJSONRPCServer.auth_required @defer.inlineCallbacks def jsonrpc_claim_send_to_address(self, claim_id, address, amount=None): """ Send a name claim to an address Usage: claim_send_to_address (<claim_id> | --claim_id=<claim_id>) (<address> | --address=<address>) [<amount> | --amount=<amount>] Options: <amount> : Amount of credits to claim name for, defaults to the current amount on the claim """ result = yield self.session.wallet.send_claim_to_address(claim_id, address, amount) response = yield self._render_response(result) defer.returnValue(response) # TODO: claim_list_mine should be merged into claim_list, but idk how to authenticate it -Grin @AuthJSONRPCServer.auth_required def jsonrpc_claim_list_mine(self): """ List my name claims Usage: claim_list_mine Returns (list) List of name claims owned by user [ { 'address': (str) address that owns the claim 'amount': (float) amount assigned to the claim 'blocks_to_expiration': (int) number of blocks until it expires 'category': (str) "claim", "update" , or "support" 'claim_id': (str) claim ID of the claim 'confirmations': (int) number of blocks of confirmations for the claim 'expiration_height': (int) the block height which the claim will expire 'expired': (bool) true if expired, false otherwise 'height': (int) height of the block containing the claim 'is_spent': (bool) true if claim is abandoned, false otherwise 'name': (str) name of the claim 'txid': (str) txid of the cliam 'nout': (int) nout of the claim 'value': (str) value of the claim }, ] """ d = self.session.wallet.get_name_claims() d.addCallback(lambda claims: self._render_response(claims)) return d @defer.inlineCallbacks def jsonrpc_claim_list(self, name): """ List current claims and information about them for a given name Usage: claim_list (<name> | --name=<name>) Returns (dict) State of claims assigned for the name { 'claims': (list) list of claims for the name [ { 'amount': (float) amount assigned to the claim 'effective_amount': (float) total amount assigned to the claim, including supports 'claim_id': (str) claim ID of the claim 'height': (int) height of block containing the claim 'txid': (str) txid of the claim 'nout': (int) nout of the claim 'supports': (list) a list of supports attached to the claim 'value': (str) the value of the claim }, ] 'supports_without_claims': (list) supports without any claims attached to them 'last_takeover_height': (int) the height of last takeover for the name } """ claims = yield self.session.wallet.get_claims_for_name(name) defer.returnValue(claims) @AuthJSONRPCServer.auth_required @defer.inlineCallbacks def jsonrpc_claim_list_by_channel(self, page=0, page_size=10, uri=None, uris=[]): """ Get paginated claims in a channel specified by a channel uri Usage: claim_list_by_channel (<uri> | --uri=<uri>) [<uris>...] [--page=<page>] [--page_size=<page_size>] Options: --page=<page> : which page of results to return where page 1 is the first page, defaults to no pages --page_size=<page_size> : number of results in a page, default of 10 Returns: { resolved channel uri: { If there was an error: 'error': (str) error message 'claims_in_channel': the total number of results for the channel, If a page of results was requested: 'returned_page': page number returned, 'claims_in_channel': [ { 'absolute_channel_position': (int) claim index number in sorted list of claims which assert to be part of the channel 'address': (str) claim address, 'amount': (float) claim amount, 'effective_amount': (float) claim amount including supports, 'claim_id': (str) claim id, 'claim_sequence': (int) claim sequence number, 'decoded_claim': (bool) whether or not the claim value was decoded, 'height': (int) claim height, 'depth': (int) claim depth, 'has_signature': (bool) included if decoded_claim 'name': (str) claim name, 'supports: (list) list of supports [{'txid': (str) txid, 'nout': (int) nout, 'amount': (float) amount}], 'txid': (str) claim txid, 'nout': (str) claim nout, 'signature_is_valid': (bool), included if has_signature, 'value': ClaimDict if decoded, otherwise hex string } ], } } """ uris = tuple(uris) if uri is not None: uris += (uri, ) results = {} valid_uris = tuple() for chan_uri in uris: try: parsed = parse_lbry_uri(chan_uri) if not parsed.is_channel: results[chan_uri] = {"error": "%s is not a channel uri" % parsed.name} elif parsed.path: results[chan_uri] = {"error": "%s is a claim in a channel" % parsed.path} else: valid_uris += (chan_uri, ) except URIParseError: results[chan_uri] = {"error": "%s is not a valid uri" % chan_uri} resolved = yield self.session.wallet.resolve(*valid_uris, check_cache=False, page=page, page_size=page_size) for u in resolved: if 'error' in resolved[u]: results[u] = resolved[u] else: results[u] = { 'claims_in_channel': resolved[u]['claims_in_channel'] } if page: results[u]['returned_page'] = page results[u]['claims_in_channel'] = resolved[u].get('claims_in_channel', []) response = yield self._render_response(results) defer.returnValue(response) @AuthJSONRPCServer.auth_required @AuthJSONRPCServer.flags(include_tip_info='-t') def jsonrpc_transaction_list(self, include_tip_info=False): """ List transactions belonging to wallet Usage: transaction_list [-t] Options: -t : Include claim tip information Returns: (list) List of transactions, where is_tip is null by default, and set to a boolean if include_tip_info is true { "claim_info": (list) claim info if in txn [{"amount": (float) claim amount, "claim_id": (str) claim id, "claim_name": (str) claim name, "nout": (int) nout}], "confirmations": (int) number of confirmations for the txn, "date": (str) date and time of txn, "fee": (float) txn fee, "support_info": (list) support info if in txn [{"amount": (float) support amount, "claim_id": (str) claim id, "claim_name": (str) claim name, "is_tip": (null) default, (bool) if include_tip_info is true, "nout": (int) nout}], "timestamp": (int) timestamp, "txid": (str) txn id, "update_info": (list) update info if in txn [{"amount": (float) updated amount, "claim_id": (str) claim id, "claim_name": (str) claim name, "nout": (int) nout}], "value": (float) value of txn } """ d = self.session.wallet.get_history(include_tip_info) d.addCallback(lambda r: self._render_response(r)) return d def jsonrpc_transaction_show(self, txid): """ Get a decoded transaction from a txid Usage: transaction_show (<txid> | --txid=<txid>) Returns: (dict) JSON formatted transaction """ d = self.session.wallet.get_transaction(txid) d.addCallback(lambda r: self._render_response(r)) return d @AuthJSONRPCServer.auth_required def jsonrpc_wallet_is_address_mine(self, address): """ Checks if an address is associated with the current wallet. Usage: wallet_is_address_mine (<address> | --address=<address>) Returns: (bool) true, if address is associated with current wallet """ d = self.session.wallet.address_is_mine(address) d.addCallback(lambda is_mine: self._render_response(is_mine)) return d @AuthJSONRPCServer.auth_required def jsonrpc_wallet_public_key(self, address): """ Get public key from wallet address Usage: wallet_public_key (<address> | --address=<address>) Returns: (list) list of public keys associated with address. Could contain more than one public key if multisig. """ d = self.session.wallet.get_pub_keys(address) d.addCallback(lambda r: self._render_response(r)) return d @AuthJSONRPCServer.auth_required @defer.inlineCallbacks def jsonrpc_wallet_list(self): """ List wallet addresses Usage: wallet_list Returns: List of wallet addresses """ addresses = yield self.session.wallet.list_addresses() response = yield self._render_response(addresses) defer.returnValue(response) @AuthJSONRPCServer.auth_required def jsonrpc_wallet_new_address(self): """ Generate a new wallet address Usage: wallet_new_address Returns: (str) New wallet address in base58 """ def _disp(address): log.info("Got new wallet address: " + address) return defer.succeed(address) d = self.session.wallet.get_new_address() d.addCallback(_disp) d.addCallback(lambda address: self._render_response(address)) return d @AuthJSONRPCServer.auth_required def jsonrpc_wallet_unused_address(self): """ Return an address containing no balance, will create a new address if there is none. Usage: wallet_unused_address Returns: (str) Unused wallet address in base58 """ def _disp(address): log.info("Got unused wallet address: " + address) return defer.succeed(address) d = self.session.wallet.get_unused_address() d.addCallback(_disp) d.addCallback(lambda address: self._render_response(address)) return d @AuthJSONRPCServer.deprecated("wallet_send") @AuthJSONRPCServer.auth_required @defer.inlineCallbacks def jsonrpc_send_amount_to_address(self, amount, address): """ Queue a payment of credits to an address Usage: send_amount_to_address (<amount> | --amount=<amount>) (<address> | --address=<address>) Returns: (bool) true if payment successfully scheduled """ if amount < 0: raise NegativeFundsError() elif not amount: raise NullFundsError() reserved_points = self.session.wallet.reserve_points(address, amount) if reserved_points is None: raise InsufficientFundsError() yield self.session.wallet.send_points_to_address(reserved_points, amount) self.analytics_manager.send_credits_sent() defer.returnValue(True) @AuthJSONRPCServer.auth_required @defer.inlineCallbacks def jsonrpc_wallet_send(self, amount, address=None, claim_id=None): """ Send credits. If given an address, send credits to it. If given a claim id, send a tip to the owner of a claim specified by uri. A tip is a claim support where the recipient of the support is the claim address for the claim being supported. Usage: wallet_send (<amount> | --amount=<amount>) ((<address> | --address=<address>) | (<claim_id> | --claim_id=<claim_id>)) Return: If sending to an address: (bool) true if payment successfully scheduled If sending a claim tip: (dict) Dictionary containing the result of the support { txid : (str) txid of resulting support claim nout : (int) nout of the resulting support claim fee : (float) fee paid for the transaction } """ if address and claim_id: raise Exception("Given both an address and a claim id") elif not address and not claim_id: raise Exception("Not given an address or a claim id") if amount < 0: raise NegativeFundsError() elif not amount: raise NullFundsError() if address: # raises an error if the address is invalid decode_address(address) result = yield self.jsonrpc_send_amount_to_address(amount, address) else: validate_claim_id(claim_id) result = yield self.session.wallet.tip_claim(claim_id, amount) self.analytics_manager.send_claim_action('new_support') defer.returnValue(result) @AuthJSONRPCServer.auth_required @defer.inlineCallbacks @AuthJSONRPCServer.flags(no_broadcast='--no_broadcast') def jsonrpc_wallet_prefill_addresses(self, num_addresses, amount, no_broadcast=False): """ Create new addresses, each containing `amount` credits Usage: wallet_prefill_addresses [--no_broadcast] (<num_addresses> | --num_addresses=<num_addresses>) (<amount> | --amount=<amount>) Returns: (dict) the resulting transaction """ if amount < 0: raise NegativeFundsError() elif not amount: raise NullFundsError() broadcast = not no_broadcast tx = yield self.session.wallet.create_addresses_with_balance( num_addresses, amount, broadcast=broadcast) tx['broadcast'] = broadcast defer.returnValue(tx) @defer.inlineCallbacks def jsonrpc_wallet_list_unspent(self): """ List unspent transaction outputs Usage: wallet_prefill_addresses Returns: (list) List of UTXOs [ { "address": (str) the output address "amount": (float) unspent amount "height": (int) block height "is_claim": (bool) is the tx a claim "is_coinbase": (bool) is the tx a coinbase tx "is_support": (bool) is the tx a support "is_update": (bool) is the tx an update "nout": (int) nout of the output "txid": (str) txid of the output }, ... ] """ unspent = yield self.session.wallet.list_unspent() for i, utxo in enumerate(unspent): utxo['txid'] = utxo.pop('prevout_hash') utxo['nout'] = utxo.pop('prevout_n') utxo['amount'] = utxo.pop('value') utxo['is_coinbase'] = utxo.pop('coinbase') unspent[i] = utxo defer.returnValue(unspent) def jsonrpc_block_show(self, blockhash=None, height=None): """ Get contents of a block Usage: block_show (<blockhash> | --blockhash=<blockhash>) | (<height> | --height=<height>) Options: <blockhash>, --blockhash=<blockhash> : hash of the block to look up <height>, --height=<height> : height of the block to look up Returns: (dict) Requested block """ if blockhash is not None: d = self.session.wallet.get_block(blockhash) elif height is not None: d = self.session.wallet.get_block_info(height) d.addCallback(lambda b: self.session.wallet.get_block(b)) else: # TODO: return a useful error message return server.failure d.addCallback(lambda r: self._render_response(r)) return d @AuthJSONRPCServer.auth_required @defer.inlineCallbacks def jsonrpc_blob_get(self, blob_hash, timeout=None, encoding=None, payment_rate_manager=None): """ Download and return a blob Usage: blob_get (<blob_hash> | --blob_hash=<blob_hash>) [--timeout=<timeout>] [--encoding=<encoding>] [--payment_rate_manager=<payment_rate_manager>] Options: --timeout=<timeout> : timeout in number of seconds --encoding=<encoding> : by default no attempt at decoding is made, can be set to one of the following decoders: 'json' --payment_rate_manager=<payment_rate_manager> : if not given the default payment rate manager will be used. supported alternative rate managers: 'only-free' Returns (str) Success/Fail message or (dict) decoded data """ decoders = { 'json': json.loads } timeout = timeout or 30 payment_rate_manager = get_blob_payment_rate_manager(self.session, payment_rate_manager) blob = yield self._download_blob(blob_hash, rate_manager=payment_rate_manager, timeout=timeout) if encoding and encoding in decoders: blob_file = blob.open_for_reading() result = decoders[encoding](blob_file.read()) blob_file.close() else: result = "Downloaded blob %s" % blob_hash response = yield self._render_response(result) defer.returnValue(response) @AuthJSONRPCServer.auth_required @defer.inlineCallbacks def jsonrpc_blob_delete(self, blob_hash): """ Delete a blob Usage: blob_delete (<blob_hash> | --blob_hash=<blob_hash) Returns: (str) Success/fail message """ if blob_hash not in self.session.blob_manager.blobs: response = yield self._render_response("Don't have that blob") defer.returnValue(response) try: stream_hash = yield self.stream_info_manager.get_stream_hash_for_sd_hash(blob_hash) yield self.stream_info_manager.delete_stream(stream_hash) except Exception as err: pass yield self.session.blob_manager.delete_blobs([blob_hash]) response = yield self._render_response("Deleted %s" % blob_hash) defer.returnValue(response) def jsonrpc_peer_list(self, blob_hash, timeout=None): """ Get peers for blob hash Usage: peer_list (<blob_hash> | --blob_hash=<blob_hash>) [<timeout> | --timeout=<timeout>] Options: <timeout>, --timeout=<timeout> : peer search timeout in seconds Returns: (list) List of contacts """ timeout = timeout or conf.settings['peer_search_timeout'] d = self.session.peer_finder.find_peers_for_blob(blob_hash, timeout=timeout) d.addCallback(lambda r: [[c.host, c.port, c.is_available()] for c in r]) d.addCallback(lambda r: self._render_response(r)) return d @defer.inlineCallbacks @AuthJSONRPCServer.flags(announce_all="-a") def jsonrpc_blob_announce(self, announce_all=None, blob_hash=None, stream_hash=None, sd_hash=None): """ Announce blobs to the DHT Usage: blob_announce [-a] [<blob_hash> | --blob_hash=<blob_hash>] [<stream_hash> | --stream_hash=<stream_hash>] [<sd_hash> | --sd_hash=<sd_hash>] Options: -a : announce all the blobs possessed by user <blob_hash>, --blob_hash=<blob_hash> : announce a blob, specified by blob_hash <stream_hash>, --stream_hash=<stream_hash> : announce all blobs associated with stream_hash <sd_hash>, --sd_hash=<sd_hash> : announce all blobs associated with sd_hash and the sd_hash itself Returns: (bool) true if successful """ if announce_all: yield self.session.blob_manager.immediate_announce_all_blobs() else: if blob_hash: blob_hashes = [blob_hash] elif stream_hash: blobs = yield self.get_blobs_for_stream_hash(stream_hash) blob_hashes = [blob.blob_hash for blob in blobs if blob.get_is_verified()] elif sd_hash: blobs = yield self.get_blobs_for_sd_hash(sd_hash) blob_hashes = [sd_hash] + [blob.blob_hash for blob in blobs if blob.get_is_verified()] else: raise Exception('single argument must be specified') yield self.session.blob_manager._immediate_announce(blob_hashes) response = yield self._render_response(True) defer.returnValue(response) @AuthJSONRPCServer.deprecated("blob_announce") def jsonrpc_blob_announce_all(self): """ Announce all blobs to the DHT Usage: blob_announce_all Returns: (str) Success/fail message """ return self.jsonrpc_blob_announce(announce_all=True) @defer.inlineCallbacks def jsonrpc_file_reflect(self, **kwargs): """ Reflect all the blobs in a file matching the filter criteria Usage: file_reflect [--sd_hash=<sd_hash>] [--file_name=<file_name>] [--stream_hash=<stream_hash>] [--claim_id=<claim_id>] [--outpoint=<outpoint>] [--rowid=<rowid>] [--name=<name>] [--reflector=<reflector>] Options: --sd_hash=<sd_hash> : get file with matching sd hash --file_name=<file_name> : get file with matching file name in the downloads folder --stream_hash=<stream_hash> : get file with matching stream hash --claim_id=<claim_id> : get file with matching claim id --outpoint=<outpoint> : get file with matching claim outpoint --rowid=<rowid> : get file with matching row id --name=<name> : get file with matching associated name claim --reflector=<reflector> : reflector server, ip address or url by default choose a server from the config Returns: (list) list of blobs reflected """ reflector_server = kwargs.get('reflector', None) lbry_files = yield self._get_lbry_files(**kwargs) if len(lbry_files) > 1: raise Exception('Too many (%i) files found, need one' % len(lbry_files)) elif not lbry_files: raise Exception('No file found') lbry_file = lbry_files[0] results = yield reupload.reflect_stream(lbry_file, reflector_server=reflector_server) defer.returnValue(results) @defer.inlineCallbacks @AuthJSONRPCServer.flags(needed="-n", finished="-f") def jsonrpc_blob_list(self, uri=None, stream_hash=None, sd_hash=None, needed=None, finished=None, page_size=None, page=None): """ Returns blob hashes. If not given filters, returns all blobs known by the blob manager Usage: blob_list [-n] [-f] [<uri> | --uri=<uri>] [<stream_hash> | --stream_hash=<stream_hash>] [<sd_hash> | --sd_hash=<sd_hash>] [<page_size> | --page_size=<page_size>] [<page> | --page=<page>] Options: -n : only return needed blobs -f : only return finished blobs <uri>, --uri=<uri> : filter blobs by stream in a uri <stream_hash>, --stream_hash=<stream_hash> : filter blobs by stream hash <sd_hash>, --sd_hash=<sd_hash> : filter blobs by sd hash <page_size>, --page_size=<page_size> : results page size <page>, --page=<page> : page of results to return Returns: (list) List of blob hashes """ if uri: metadata = yield self._resolve_name(uri) sd_hash = utils.get_sd_hash(metadata) blobs = yield self.get_blobs_for_sd_hash(sd_hash) elif stream_hash: try: blobs = yield self.get_blobs_for_stream_hash(stream_hash) except NoSuchStreamHash: blobs = [] elif sd_hash: try: blobs = yield self.get_blobs_for_sd_hash(sd_hash) except NoSuchSDHash: blobs = [] else: blobs = self.session.blob_manager.blobs.itervalues() if needed: blobs = [blob for blob in blobs if not blob.get_is_verified()] if finished: blobs = [blob for blob in blobs if blob.get_is_verified()] blob_hashes = [blob.blob_hash for blob in blobs] page_size = page_size or len(blob_hashes) page = page or 0 start_index = page * page_size stop_index = start_index + page_size blob_hashes_for_return = blob_hashes[start_index:stop_index] response = yield self._render_response(blob_hashes_for_return) defer.returnValue(response) def jsonrpc_blob_reflect_all(self): """ Reflects all saved blobs Usage: blob_reflect_all Returns: (bool) true if successful """ d = self.session.blob_manager.get_all_verified_blobs() d.addCallback(reupload.reflect_blob_hashes, self.session.blob_manager) d.addCallback(lambda r: self._render_response(r)) return d def jsonrpc_routing_table_get(self): """ Get DHT routing information Usage: routing_table_get Returns: (dict) dictionary containing routing and contact information { "buckets": { <bucket index>: [ { "address": (str) peer address, "node_id": (str) peer node id, "blobs": (list) blob hashes announced by peer } ] }, "contacts": (list) contact node ids, "blob_hashes": (list) all of the blob hashes stored by peers in the list of buckets, "node_id": (str) the local dht node id } """ result = {} data_store = deepcopy(self.session.dht_node._dataStore._dict) datastore_len = len(data_store) hosts = {} if datastore_len: for k, v in data_store.iteritems(): for value, lastPublished, originallyPublished, originalPublisherID in v: try: contact = self.session.dht_node._routingTable.getContact( originalPublisherID) except ValueError: continue if contact in hosts: blobs = hosts[contact] else: blobs = [] blobs.append(k.encode('hex')) hosts[contact] = blobs contact_set = [] blob_hashes = [] result['buckets'] = {} for i in range(len(self.session.dht_node._routingTable._buckets)): for contact in self.session.dht_node._routingTable._buckets[i]._contacts: contacts = result['buckets'].get(i, []) if contact in hosts: blobs = hosts[contact] del hosts[contact] else: blobs = [] host = { "address": contact.address, "node_id": contact.id.encode("hex"), "blobs": blobs, } for blob_hash in blobs: if blob_hash not in blob_hashes: blob_hashes.append(blob_hash) contacts.append(host) result['buckets'][i] = contacts if contact.id.encode('hex') not in contact_set: contact_set.append(contact.id.encode("hex")) result['contacts'] = contact_set result['blob_hashes'] = blob_hashes result['node_id'] = self.session.dht_node.node_id.encode('hex') return self._render_response(result) @defer.inlineCallbacks def jsonrpc_get_availability(self, uri, sd_timeout=None, peer_timeout=None): """ Get stream availability for lbry uri Usage: get_availability (<uri> | --uri=<uri>) [<sd_timeout> | --sd_timeout=<sd_timeout>] [<peer_timeout> | --peer_timeout=<peer_timeout>] Options: <sd_timeout>, --sd_timeout=<sd_timeout> : sd blob download timeout <peer_timeout>, --peer_timeout=<peer_timeout> : how long to look for peers Returns: (float) Peers per blob / total blobs """ def _get_mean(blob_availabilities): peer_counts = [] for blob_availability in blob_availabilities: for blob, peers in blob_availability.iteritems(): peer_counts.append(peers) if peer_counts: return round(1.0 * sum(peer_counts) / len(peer_counts), 2) else: return 0.0 def read_sd_blob(sd_blob): sd_blob_file = sd_blob.open_for_reading() decoded_sd_blob = json.loads(sd_blob_file.read()) sd_blob_file.close() return decoded_sd_blob resolved_result = yield self.session.wallet.resolve(uri) if resolved_result and uri in resolved_result: resolved = resolved_result[uri] else: defer.returnValue(None) if 'claim' in resolved: metadata = resolved['claim']['value'] else: defer.returnValue(None) sd_hash = utils.get_sd_hash(metadata) sd_timeout = sd_timeout or conf.settings['sd_download_timeout'] peer_timeout = peer_timeout or conf.settings['peer_search_timeout'] blobs = [] try: blobs = yield self.get_blobs_for_sd_hash(sd_hash) need_sd_blob = False log.info("Already have sd blob") except NoSuchSDHash: need_sd_blob = True log.info("Need sd blob") blob_hashes = [blob.blob_hash for blob in blobs] if need_sd_blob: # we don't want to use self._download_descriptor here because it would create a stream try: sd_blob = yield self._download_blob(sd_hash, timeout=sd_timeout) except Exception as err: response = yield self._render_response(0.0) log.warning(err) defer.returnValue(response) decoded = read_sd_blob(sd_blob) blob_hashes = [blob.get("blob_hash") for blob in decoded['blobs'] if blob.get("blob_hash")] sample = random.sample(blob_hashes, min(len(blob_hashes), 5)) log.info("check peers for %i of %i blobs in stream", len(sample), len(blob_hashes)) availabilities = yield self.session.blob_tracker.get_availability_for_blobs(sample, peer_timeout) mean_availability = _get_mean(availabilities) response = yield self._render_response(mean_availability) defer.returnValue(response) @defer.inlineCallbacks @AuthJSONRPCServer.flags(a_arg='-a', b_arg='-b') def jsonrpc_cli_test_command(self, pos_arg, pos_args=[], pos_arg2=None, pos_arg3=None, a_arg=False, b_arg=False): """ This command is only for testing the CLI argument parsing Usage: cli_test_command [-a] [-b] (<pos_arg> | --pos_arg=<pos_arg>) [<pos_args>...] [--pos_arg2=<pos_arg2>] [--pos_arg3=<pos_arg3>] Options: -a, --a_arg : a arg -b, --b_arg : b arg <pos_arg2>, --pos_arg2=<pos_arg2> : pos arg 2 <pos_arg3>, --pos_arg3=<pos_arg3> : pos arg 3 Returns: pos args """ out = (pos_arg, pos_args, pos_arg2, pos_arg3, a_arg, b_arg) response = yield self._render_response(out) defer.returnValue(response) def loggly_time_string(dt): formatted_dt = dt.strftime("%Y-%m-%dT%H:%M:%S") milliseconds = str(round(dt.microsecond * (10.0 ** -5), 3)) return urllib.quote_plus(formatted_dt + milliseconds + "Z") def get_loggly_query_string(installation_id): base_loggly_search_url = "https://lbry.loggly.com/search#" now = utils.now() yesterday = now - utils.timedelta(days=1) params = { 'terms': 'json.installation_id:{}*'.format(installation_id[:SHORT_ID_LEN]), 'from': loggly_time_string(yesterday), 'to': loggly_time_string(now) } data = urllib.urlencode(params) return base_loggly_search_url + data def report_bug_to_slack(message, installation_id, platform_name, app_version): webhook = utils.deobfuscate(conf.settings['SLACK_WEBHOOK']) payload_template = "os: %s\n version: %s\n<%s|loggly>\n%s" payload_params = ( platform_name, app_version, get_loggly_query_string(installation_id), message ) payload = { "text": payload_template % payload_params } requests.post(webhook, json.dumps(payload)) def get_lbry_file_search_value(search_fields): for searchtype in FileID: value = search_fields.get(searchtype, None) if value is not None: return searchtype, value raise NoValidSearch('{} is missing a valid search type'.format(search_fields)) def iter_lbry_file_search_values(search_fields): for searchtype in FileID: value = search_fields.get(searchtype, None) if value is not None: yield searchtype, value def get_blob_payment_rate_manager(session, payment_rate_manager=None): if payment_rate_manager: rate_managers = { 'only-free': OnlyFreePaymentsManager() } if payment_rate_manager in rate_managers: payment_rate_manager = rate_managers[payment_rate_manager] log.info("Downloading blob with rate manager: %s", payment_rate_manager) return payment_rate_manager or session.payment_rate_manager